A real-time modular rack for guitar, voice and synthesis — patch it like hardware, play it like an instrument.
SignalPatch turns an audio interface into a patchable rack: drag cables between modules while the sound keeps running, stomp any effect in or out of the chain, feed real Neural Amp Modeler profiles, vocode and autotune a voice, sequence a synth and a drum machine, and bounce ideas onto a virtual 4-track — all in one native C++/JUCE application built around a real-time-safe render graph.
SignalPatch is being built toward a specific destination: a handheld
device — an ASUS ROG Ally class machine — as a self-contained guitar
patching studio, voice processor and synthesizer. Plug a guitar or a mic
into a small interface, pick up the handheld, and the whole rack is an
instrument you hold: stomp switches on triggers, macro knobs on sticks,
patching on the touchscreen. Every phase of the roadmap
walks toward that. The dedicated-machine build guide (stripped-down Linux,
boot-to-instrument kiosk, --kiosk --unmute flags, latency budgeting) lives
in docs/APPLIANCE.md.
- 36 modules across seven families: utility (gain, 4-ch mixer, crossfade) · effects (distortion, SVF filter, delay, reverb, chorus, phaser, tremolo, bitcrusher, ring mod, pitch shifter, granular cloud) · neural (Neural Amp head and Neural Pedal — any .nam capture, stepped through your model folder like a pedal library, each showing its own per-block cost) · voice (vowel/formant filter, 12-band vocoder, autotune) · instruments (mono synth, Karplus-Strong pluck, noise, drum machine, live-input sampler, 60 s 4-track tape with varispeed) · dynamics (compressor, limiter, noise gate, Feedback Guard) · control (LFO, random S&H, envelope follower, 8-step sequencer, macro, FFT spectral follower, and a Script module that compiles a per-sample math expression on the fly).
- Neural Amp Modeler, for real. Load any
.namprofile; parsing happens on a worker thread and the model swaps into the audio path atomically. The model path is saved with the patch, and a benchmark harness qualifies any folder of captures against the 64-sample deadline (SIGNALPATCH_BENCH_NAM_DIR=... ./signalpatch_tests). - Rack interaction. Accent-coloured face plates with rails, screws and activity LEDs; true-bypass stomp footswitches; cables that sag, glow with signal level and carry flow pulses toward their destination; right-click patching; a signal-aware inspector.
- Feedback as a feature, safely. Cycles are legal only through a Feedback Guard (causal delay → DC block → finite check → hard ceiling with runaway latching). Outputs carry a fixed safety ceiling; loaded patches come back muted and fade in deliberately.
- Everything is modulatable. Every knob has a mod socket; any control signal (LFO, envelope, sequencer, spectral band, macro, MIDI later) can drive any parameter.
- Session honesty. Versioned JSON patches, autosave recovery on launch, device/channel identity preserved even when the interface changes.
The audio callback allocates nothing, locks nothing and never blocks — and
that is tested, not asserted: a global allocation trap arms around a
worst-case graph (every module, guarded feedback, live NAM inference) in the
test suite, a 30-minutes-of-audio soak runs the same graph, and the suite
passes ASan+UBSan. Graph edits compile an immutable snapshot on the message
thread and swap at a block boundary; retired snapshots are reclaimed off the
audio thread. Details: docs/ARCHITECTURE.md,
docs/REALTIME_SAFETY.md,
docs/PRODUCTION_READINESS.md.
Requirements: CMake ≥ 3.22, a C++20 compiler, Ninja/Make/VS2022. JUCE 8.0.13
and NeuralAmpModelerCore are fetched automatically when not found locally
(-DSIGNALPATCH_ALLOW_JUCE_FETCH=OFF disables the JUCE fallback;
-DSIGNALPATCH_ENABLE_NAM=OFF builds without the neural amp).
sudo apt install ninja-build libasound2-dev libjack-jackd2-dev \
libx11-dev libxrandr-dev libxinerama-dev libxcursor-dev libxext-dev \
libfreetype6-dev libfontconfig1-dev
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=RelWithDebInfo
cmake --build build --parallel
ctest --test-dir build --output-on-failure
./build/signalpatch_artefacts/RelWithDebInfo/SignalPatch # ALSA/JACK
pw-jack ./build/signalpatch_artefacts/RelWithDebInfo/SignalPatch # PipeWireflatpak install flathub org.flatpak.Builder org.freedesktop.Sdk//24.08
flatpak run org.flatpak.Builder --user --install --force-clean \
build-flatpak packaging/flatpak/io.github.willbearfruits.SignalPatch.yml
flatpak run io.github.willbearfruits.SignalPatchFrom a Visual Studio (2022 or newer) developer shell:
cmake -S . -B build -A x64
cmake --build build --config Release --parallel
ctest --test-dir build -C Release --output-on-failure
.\build\signalpatch_artefacts\Release\SignalPatch.exe(CMake picks your installed Visual Studio; add -G "Visual Studio 17 2022"
to pin a specific one.)
WASAPI is on by default; add -DSIGNALPATCH_ENABLE_ASIO=ON for ASIO (brings
the Steinberg ASIO SDK licence terms into scope).
A patch file can be passed on the command line:
SignalPatch demo-pedalboard.signalpatch (loads muted; press MUTED to fade
in — the same applies to the restored last session).
The Neural Amp module builds against
NeuralAmpModelerCore
(pinned commit, fetched automatically, with a tiny guard patch from
packaging/patches/ for toolchains whose libstdc++ predates
std::atomic<std::shared_ptr>). Point SIGNALPATCH_NAM_CORE_DIR at a local
checkout to skip the fetch. Get .nam profiles from
ToneHunt and load them with the module's
LOAD MODEL button; performance qualification for heavy models at small
buffers is roadmap item 0.4.
| Doc | What it is |
|---|---|
docs/ROADMAP.md |
The ordered plan, 0.2 → handheld instrument |
docs/APPLIANCE.md |
Building the dedicated handheld rig (ROG Ally + Zoom F4) |
docs/PRODUCT.md |
Product specification |
docs/ARCHITECTURE.md |
Engine design: document → compiler → snapshot |
docs/REALTIME_SAFETY.md |
The audio-callback contract (a release gate) |
docs/NAM_ROADMAP.md |
Neural amp staging and gates |
docs/PRODUCTION_READINESS.md |
Which verification gates are closed vs. open |
SignalPatch is released under the GNU AGPL-3.0 (see LICENSE), matching JUCE 8's open-source licence tier. NeuralAmpModelerCore is MIT; enabling ASIO additionally involves Steinberg's SDK licence.